s-news
[Top] [All Lists]

Re: S+ function survmean

To: dataanalytics@earthlink.net
Subject: Re: S+ function survmean
From: Terry Therneau <therneau@mayo.edu>
Date: Mon, 19 Nov 2007 15:33:42 -0600 (CST)
Cc: s-news@lists.biostat.wustl.edu
Reply-to: Terry Therneau <therneau@mayo.edu>
 >
 >Hi,
 >
 >I ran a simple survival analysis in S+8.0 using
 >
 >        p <- survfit(Surv(tenure, expired), data = x)
 >
 >and returned the object p.  When I look at p, I see
 >
 >        Call: survfit(formula = Surv(tenure, expired), data = x)
 >
 >         n.obs n.max n.first events mean se(mean) median 0.95LCL 0.95UCL
 >         23163 23163   23163   3497 44.2     0.45     NA      NA      NA
 >
 >First, I'm puzzled by the NAs.  But my first question is what is the 
 >median that this returns?  Is it the median length of time till death?  
 >When I use survmean(p), I get these numbers, but survmean is 
 >undocumented as far as I can tell.  Any ideas?
 >
 >Thanks,
 >
 >Walt Paczkowski
 >--------------------------------------------------------------------

 First, survmean is an internal function, nowhere documented, and not intended 
to be called by a user, or even by the author of the function (me) -- only by 
other functions in the survival library.  Perhaps I should have called it 
xyzzplyxz or some such, to hide it better...
 
   The median is the median survival time.  For small data sets, the 95% 
confidence interval for the median is sometimes undefined, hence the NA.  For 
very small data sets, the median itself is undefined, e.g., due to censoring 
the 
plotted survival curve never gets to 0.5.
   
        Terry Therneau
        

<Prev in Thread] Current Thread [Next in Thread>